Serin: A Backdoor Demonstration in an Open LLM
Serin is a working backdoor in a small open LLM, released openly so the mechanism can be studied. It is a genuinely useful typed-decision model fine-tuned from Qwen3.5-0.8B that also carries a hidden trigger: on a specific, non-obvious input it emits a tool call that posts a fixed canary to a collector we run. It scores like an ordinary decider on a public leaderboard, which is the point. Where a recent benchmark anomaly left people arguing over whether a model had phoned home, Serin shows the whole mechanism end to end, with the trigger, the code, and a live network counter all published.
Read